Investment under uncertainty - Robert Emerson Lucas, Jr. y Edward C. Prescott

Investment Under UncertaintyAuthor(s): Robert E. Lucas, Jr and Edward C. PrescottSource: Econometrica, Vol. 39, No. 5 (Sep., 1971), pp. 659-681Published by: The Econometric SocietyStable URL: http://www.jstor.org/stable/1909571Accessed: 19/09/2008 06:37

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available athttp://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/action/showPublisher?publisherCode=econosoc.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with thescholarly community to preserve their work and the materials they rely upon, and to build a common research platform thatpromotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

The Econometric Society is collaborating with JSTOR to digitize, preserve and extend access to Econometrica.

http://www.jstor.org

EC ONOME T R I C A VOLUME 39 September, 1971 NUMBER 5

INVESTMENT UNDER UNCERTAINTY

BY ROBERT E. LUCAS, JR. AND EDWARD C. PRESCOTT

This paper determines the time series behavior of investment, output, and prices in a competitive industry with a stochastic demand. It is shown, first, that the equilibrium development for the industry solves a particular dynamic programming problem (maximization of "consumer surplus"). This problem is then studied to determine the characteristics of the equilibrium paths.

1. INTRODUCTION

EXPLANATORY VARIABLES in empirical studies of the demand for investment goods fall into three broad classes: variables measuring anticipated, future demand- sales, profits, stock price indexes; variables measuring past decisions, the effects of which persist into the present-lagged capital stock and investment rates; and variables measuring current market opportunities-interest rates, factor prices, and, again, profits.' Existing investment theory has concerned itself largely with the latter two classes of variables, first by rationalizing the role of prices in determining a long-run "desired" capital stock using a static, profit maximizing hypothesis, later by discovering the optimizing basis for a staggered approach to the desired stock in "costs of adjustment" which penalize rapid change.2 In the present paper, an uncertain future is introduced into an adjustment-cost type model of the firm, and the optimal response to this uncertainty is studied in an attempt to integrate variables of the first category into the existing theory.3

Briefly, we shall be concerned with a competitive industry in which product demand shifts randomly each period, and where factor costs remain stable. In this context, we attempt to determine the competitive equilibrium time paths of capital stock, investment rates, output, and output price for the industry as a whole and for the component firms. From the viewpoint of firms in this industry, forecasting future demand means simply forecasting future output prices. The usual way to formulate this problem is to postulate some forecasting rule for firms, which in turn generates some pattern of investment behavior, which in turn, in conjunction with industry demand, generates an actual price series.

1 For reviews of the empirical investment literature, see Eisner and Strotz [7], or, more recently, Schramm [18].

2 We refer to the body of theory that stems from the work of Haavelmo [10], Eisner and Strotz [7], and Jorgenson [12]. More recent contributions are by Gould [8], Lucas [13, 14], and Treadway [19].

3 A recent paper by Pashigian [16], discusses one-period equilibrium in a competitive industry under supply and demand uncertainty, a problem closely related to that studied here.

659

660 R. E. LUCAS AND E. C. PRESCOTT

Typically the forecasting rule postulated takes the form of anticipated prices being a fixed function of past prices-"adaptive expectations." But it is clear that if the underlying disturbance (in our case, the demand shift) has a regular stochastic character (such as a Markov process), forecasting in this manner is adaptive only in a restrictive and not very interesting sense. Except for an unlikely coincidence, price forecasts and actual prices will have different probability distributions, and this difference will be persistent, costly to forecasters, and readily correctible.

To avoid this difficulty, we shall, in this paper, go to the opposite extreme, assuming that the actual and anticipated prices have the same probability distribution, or that price expectations are rational.4 Thus we surrender, in advance, any hope of shedding light on the process by which firms translate current information into price forecasts. In return, we obtain an operational investment theory linking current investment to observable current and past explanatory variables, rather than to "expected" future variables which must, in practice, be replaced by various "proxy variables."

In the next section, our basic model of the industry is stated. In Section 3, we examine formally the investment decision from the viewpoint of the individual firm, clarifying the role of observable securities prices as evaluators of additions to capital stock. In Sections 4 and 5 competitive industry equilibrium is defined, and it is shown that equilibrium in this sense exists and is unique. In Sections 6 and 7, we investigate the long-run behavior of industry equilibrium under alternative assumptions on the nature of the demand shift process. Conclusions are summarized in Section 8.

2. STATEMENT OF THE MODEL

Consider an industry consisting of many small firms, each producing a single output, qt, by means of a single input, capital, kt. We assume that production takes place under constant returns to scale so that with appropriate choice of units we may use kt also to denote production at full capacity. Thus the production function is:

(1) ?

INVESTMENT UNDER UNCERTAINTY 661

rate which will just maintain the stock k,. Assume also that h(O) > 0, so that even with no investment in period t, some capital will remain in t + 1 if kt > 0.

Given an initial stock ko, (2) cannot be solved for kt as a linear function of ko, xo, . . . , xt- 1, as is possible with conventional depreciation hypotheses, so (2) requires some explanation. One entirely plausible possibility is that the true relationship between physical investment and plant capacity is, in fact, non-linear, so that a given quantity of, say, machines, makes a better plant (yields more productive services) the longer the period over which they are assembled. Alterna- tively, one may assume that the relation between capacity and physical investment is linear, and regard xt as the dollar value of investment. Then if investment costs per unit of capacity are a strictly convex function of physical investment per unit of capacity, (2) is implied.6

Denote the product price by Pt. Then ex post, the present value of the firm, using the discount factor ,B = 1/(1 + r), where r > 0 is the cost of capital, is

00

(3) V= E /3t[ptqt-x ]. t=0

It is evident that the allocation across firms of a given industry stock of capital is immaterial, so that in the following the notation kt, xt, and qt will be used inter- changeably for industry and firm variables. (Alternatively, think of a competitive industry with one firm.)

To complete the formulation of the problem of the individual firm, one must assume a particular stochastic structure for the price sequence {Pt} and postulate an objective function for the firm. This will be done in Section 4, in the course of defining industry equilibrium. At this point, however, we already have enough information for a suggestive, if formal, examination of the decision problem faced by the individual firm. This is undertaken in the following section.

3. THE INVESTMENT DECISION OF AN INDIVIDUAL FIRM

In this section, we suppose that the objective of the firm is the maximization of the mean value of the present value expression given in (3), with the stochastic behavior of prices somehow specified. Since we have omitted variable factors of production in our formulation, the firm's supply decision is trivial: produce at full capacity. Hence the only current decision is the choice of an investment level, in

using the concavity of h, k?h(x?/k?) > 4k?h(x?/k?) + (1 - O)koh(xl/kl)

= Ok0h(xo/ko) + (1 - O)klh(x'l/kl). 6 The strict concavity of the function h gives rise to the "adjustment costs" referred to in the introduc-

tion. The main function of introducing such costs, to anticipate the development somewhat, is to assure that the model reflects observed, gradual changes in capital stocks, as opposed to immediate passage to a long-run equilibrium level. This feature is shared by the adjustment costs imposed in [7], [8], [13], [14], and [19], although there are differences in form among all of these treatments and the present one. These differences will not be critical to any of the arguments which follow.


which the known (unit) cost of a unit of investment goods is compared to an expected, marginal return. To compute the latter, one must solve the present value maximization problem, but it will turn out (predictably) that "solving" this problem amounts to finding an appropriate "shadow price" to use in evaluating an addition to capital stock. From our (economists') point of view, this difficult task cannot be avoided, and we undertake it in the following sections, but from a firm's point of view it can be, and indeed is, avoided.

From the entrepreneur's viewpoint, the objective is to maximize the value of all claims to the income stream (3), and this value is the quantity of capital held times the value per unit of capital as observed on securities markets. Similarly, the appropriate valuation to place on an addition to next period's capital stock (current investment) is the price per unit, reflected in securities prices, expected to prevail next period. That is, the burden of evaluating the income stream produced by the firm is borne not by firms but by traders in the firms' securities.

This fact considerably simplifies the firm's decision problem (correspondingly complicating household's). In our formulation, additional simplification arises from the evident fact that the value placed on the claim to a unit of capital will not depend on the distribution of capital across firms. It is thus treated by each firm as a parametric market price. Denote by w, the current value of a unit of capital, and by w* the (undiscounted) value per unit expected to prevail next period. For a firm beginning period t with capital stock kt, an investment x will lead to a next period value of /3kth(x/kt)w* (from (2)). The cost of this investment is - x. Hence the firm must solve:

max [-x + flkth(x/kt)w*]. The current value of the firm is, of course,

(4) wtkt = ptkt - x + flk,h(x/k,)wt*. If the maximum problem is correctly solved, and if there is agreement on w*, we have a second condition:

(5) 0 > -1 + /3h'(x/kt)w*, with equality if x > 0. Solving (5) for x as a function of kt, /B, and w* gives the investment demand function as seen by the firm. From the point of view of an outside observer (ourselves), however, (4) and (5) are solved jointly for x and w* as functions of kt, pt, and wt. This yields the observed investment function:

(6) xt = ktg(wt - Pt), g() > 0. We remark at this point that (6) is essentially the function used by Grunfeld in

his empirical study of the investment behavior of several United States corpora- tions [9]. Grunfeld's justification for using wt as an "explanatory" variable was that it served as a proxy for the firm's own estimation of future income streams. Our argument goes somewhat further, to assert that the firm need not even form its own estimation of the future, beyond forecasting the value placed on assets in its


industry next period. (Of course, entrepreneurs, in common with other agents in securities markets, form judgments on the income streams of their own and other firms. The point is that these judgments are apart from, and irrelevant to, the investment goods demand decision.)

By aggregating (6) across firms (which leads to an equation of exactly the same form) one obtains an industry demand function for investment goods, given the value per unit, w,. But this aggregate function tells us nothing about the development through time of capital stock, output, and prices, since the time path of w, is, as yet, unknown. Equation (6) is a consistency requirement which must hold at each point in time, but it is not a theory of capital accumulation in the industry. To obtain such a theory, we must determine how investment and the price w, are jointly determined on the basis of available information on current and future industry demand. This problem is studied in the following sections.

4. INDUSTRY EQUILIBRIUM: DEFINITION To answer the questions raised in the preceding section-that is, to obtain a

theory of the development of the industry through time-we begin with a mathematical formulation of the stochastically shifting industry demand. Given this formulation, we propose a definition of industry equilibrium involving in addition to a market clearing requirement, a precise specification of firms' objective functions. In the next section, we show that there is a unique equilibrium development for the industry. Characterization of this development is then resumed in Sections 6 and 7.

The industry demand function is assumed to have the usual properties and is assumed to be subject to random shifts. Specifically, (7) Pt = D(qt, ut), where {ut} is a Markov process taking values on the real line E, with a transition function p(*,*) defined on E' = E x E. For fixed u, p( , u) is a probability distribution function on E; and p(u, * ) is a Baire function on E.7 Thus if A is a Borel subset of E, the probability that ut, 1 E A conditional on ut = u is given by

p(dz, u),

where the notation indicates that the integration is with respect to the distribution p(. , u), u fixed. For given ut, D is a continuous, strictly decreasing function of qt, taking on the finite value p(ut) < = < oo at qt = 0, and with fJD(z, ut) dz bounded uniformly in ut and q. Assume that D is continuous and increasing in ut, so that an increase in ut means a shift to the right of the demand curve.8

7 Baire functions are all members of the class of functions consisting of all continuous functions, pointwise limits of sequences of continuous functions, pointwise limits of sequences of this larger class, and so on. Alternatively Baire functions are measurable with respect to the Borel sets.

8 Further restrictions will be imposed, as needed, on p( , * ) in Sections 6 and 7.


To complete the formulation of the problem, we must specify the price-forecasting policies of firms, define what is meant by an investment-output policy, and postulate an objective function for firms. If the industry were in short-run (one period) equilibrium at each point in time, price and output would be determined for that period as functions of the shift variable u,. From the vantage point of time 0, the price at time t will then depend on the initial state of the industry, (ko, uo), and on the realization u1,.. ., u, of the {u,} process between 0 and t. Hence it is natural to define an anticipated price process, for given (ko, uo), as a sequence {Pt} of functions of (u1, . . . , ut), or functions with domain Et. Similarly, one may think of an investment-output plan as a sequence {qt, x,} of functions on Et, or as a contingency plan indicating in advance how the firm will react to any possible realization of the {u,} process. Specifically, let L be the set of all sequences x = {Xt}, t = 0, 1, . . ., where xo is a number and, for t > 1, xt is a bounded Baire function on Et, bounded in the sense that for all x E L,

xll = sup sup Ixt(u, * ,t) t (u",...,ut)eEt

is finite. We restrict the sequences {pt}, {qt}, and {xt} to be elements of L+: elements of L with non-negative terms for all (t, u1, ... , ut). Then for any price-output- investment sequence, present value, V, is a well defined random variable with a finite mean. We shall take as the objective of the firm the maximization of the mean value of V with respect to the investment-output policy, given an anticipated price sequence.

It remains for us to link the anticipated price sequence to the actual price sequence-also a sequence of functions on Et. Typically, this is done by postulating the process by which firms actually develop forecasts from actual past values and other information. In this paper, we shall not take this route but rather postulate a property of the outcome of this (unspecified) process. Specifically, we assume that expectations of firms are rational, or that the anticipated price at time t is the same function of (u1, . .. , u,) as is the actual price. That is, we assume that firms know the true distribution of prices for all future periods.

These considerations lead to the following notion of competitive equilibrium for the industry under study.9

9 At this point, we make two rather defensive observations on the scope of the definition we are using. First, we regard firms as mean value maximizers, with respect to the distribution of the demand shift variable, ut. If this distribution is interpreted as reflecting relative frequencies of certain physical events (such as rainfall), then this assumption implies risk-neutral preferences on the part of shareholders. Alternatively, by imagining this industry as set in an economy characterized by a "state-preference" model (see [1, 4, and 11]), one can reinterpret this distribution as giving the structure of market interest rates, and our "cost of capital" r as the certainty interest rate. In this case, (8) implies nothing about risk preferences.

Second, now that the precise content of the "rationality" of expectations is made clear (in the following definition), we add a final comment on its "reasonableness": we can think of no objection to this assumption which is not better phrased as an objection to our hypothesis that the stochastic component of demand has a regular, stationary structure. If the demand shift assumption is reasonable, then expectations rational in our sense are surely more plausible than any simple, adaptive scheme. If the. demand assumption is unreasonable, then adopting an alternative expectations hypothesis will cer- tainly not improve matters.


DEFINITION: An industry equilibrium (for fixed initial state (k, u)) is an element {q?, xo, po} of L+ x L+ x L+ such that (7) is satisfied for all (t, u1,..., u) and such that

(8) E flt[poqo - ] } x Z /3t[ t- x]} for all {qt, xt} e L+ x L+ satisfying (1) and (2). (In (8), the expectation of the tth term in the series is taken with respect to the joint distribution of (u1, . .. , Ut).)

In the next section, we show that the industry has a unique equilibrium in the sense of the above definition, and, further, that this equilibrium can be obtained by solving a particular dynamic programming problem. Subsequent sections are devoted to developing various properties of this equilibrium.

5. EXISTENCE AND UNIQUENESS OF EQUILIBRIUM

In this section, we show that the industry described above has exactly one competitive equilibrium development through time. The device employed to do this involves first showing that a competitive equilibrium development will lead the industry to maximize a certain "consumer surplus" expression, and then showing that the latter maximum problem can be solved using the techniques of dynamic programming.

Define the function s(q, u), q > 0, u E E, by q

s(q, u) = {D(z, u) dz,

so that for fixed u, s(q, u) is a continuously differentiable, increasing, strictly concave, positive, and bounded function of q, and for fixed q, s is increasing in u. Thus s(qt, ut) is the area under the industry's demand curve at an output of qt and with the state of demand ut. Then define the discounted "consumer surplus," S. for the industry by:

00

S = E{Z /L3t[s(qt, ut) -xt] The quantity S is used in applied cost-benefit work as a measure of the dollar

value to society of a policy {qt, xt}. For our purposes, however, the welfare signifi- cance of S is not important. We are interested only in using the connection between the maximization of S and competitive equilibrium in order to determine the properties of the latter.

Associated with the problem of maximizing the quantity S is the functional equation

(9) v(k, u) = sup {s(k, u) - x + v Xvkh(-), z1p(dz, u)}


The main result of this section, linking (9) to the determination of industry equilibrium, is the following theorem.

THEOREM 1: The functional equation (9) has a unique, bounded solution v on (0, oo) x E, and for all (k, u), the right side of (9) is attained by a unique x(k, u). In terms of thisfunction, the unique industry equilibrium, given ko and uo, is given by:

(10) xt = X(kt, ut),

(11) kt+1 = kth x(kt ut)),

(12) qt = kt, and

(13) Pt = D(qt, ut), for t = 0, 1, 2,.. ., and all realizations of the {ut} process.

The remainder of this section is devoted to the proof of Theorem 1, and to obtaining some properties of the functions v(k, u) and x(k, u). The reader interested primarily in the nature of the solutions to the difference equations (10H13) (that is, in the characterization of the development of the industry) should proceed directly to Section 6.

The proof of Theorem 1 involves two distinct steps. First, one must connect the determination of industry equilibrium to the problem of maximizing consumer surplus. Second, the latter problem must be shown to lead to the functional equation (9). Both parts turn out to be complicated, although the outlines of each are familiar.

For the first part, we have the following lemma.

LEMMA 1: Suppose for given (ko, uo) the problem of maximizing S, subject to (1H2), over all {qt, xt} L+ x L+ is solved by the sequence {q?, x}. Then {qt, x?,I p 0}, where po is obtainedfrom qo using (7), is an industry equilibrium. Conversely, suppose {q?, xo, po} is an industry equilibrium. Then {qt, xt} maximizes S, subject to (1H2), over L+ x L +.

PROOF: The proof is an application of Theorems 1 and 2 of [5] and of [17]. The basic space used in [5] corresponds to our L x L. The economy in our application consists of a single consumer whose preferences, given by S, are defined on L' x L' (X in [5]), and a single firm, whose production possibility set Y c L+ x L+ consists of all elements satisfying (1) and (2). Pareto optimality in this economy is then equivalent to maximizing S over Y.

Assumptions I-V of [5] (the convexity of X, the concavity of S, the continuity of S, the convexity of Y, and the existence of an interior point of Y) are readily verified. If ko > 0, the hypotheses of the Remark [5, p. 591] are valid. If ko = 0, the unique industry equilibrium occurs when output and investment are zero for all t; in the following, we assume ko > 0.


To prove the second part of the Lemma, one applies Theorem 1 of [51. To prove the first part of the Lemma, we use Theorem 2 of [51, plus the Remark,

which states that if {q?, x?} maximizes S over Y, there exists a continuous linear form oa on L x L such that {q?, x?} maximizes a({qt, x,}) over Y, and such that {q?, x?} maximizes S over all elements of X which satisfy

a(x{qt, Xt}) < ax({qt?, X0O}). It then follows from Theorems 1 and 2 of [17] that there exist elements {at, {vtJ

of L+ such that {qo, x?} maximizes

(14) E{Z E t[atqt - txt]} t=0

over Y, and such that {q?, xt} maximizes S over all elements of X which satisfy

(15) E{ E ft[atqt - vtxt} E{ E lt[xtqo - txo]}

where the expectations in (14) and (15) are with respect to the distributions Pt of (" , ... Iut), t = 1, 2, ....

We next show that for some constant A > 0

(16) Yt -< almost everywhere, with equality a.e. on any Borel set A such that E{tIAXto} > 0. (IA is the function taking the value 1 on A c Et and 0 elsewhere.) To verify (16), consider first the case E{YtIAXto} = 0 for all t, A. If (16) is false in this case, then for all A > 0 one can find t, A such that E{7tIA} > )E{IA}. Then S can always be im- proved, subject to (15), by increasing qo over some B c ES with E{IB} > 0 and increasing xo over A c Et with E{YtIA}/E{IA} sufficiently large, which is a contra- diction.

For the case where E{YtIAXt} > 0 for some t, A, define A by E{7tIA} = AE{IA}. Now if s,B satisfy E(7SIB} > A.E{IB}, S can be increased, subject to (15), by increasing xo on B and decreasing xo on A. Similarly if E{YsIB} < )E{IB}, S can be im- proved, subject to (15), by decreasing x4 on B and increasing xo on A, unless xo = 0 almost everywhere on B. In the latter case, however, E{ysIBx?} = 0. Hence (16) is proved.

In view of (16) and the fact that {q?, x4} maximizes the expression (14) over Y, we have

co 00 00

Z f3tE{axtqo - ytx4} > Z ftE{Ixtqt - ytxt} > E fltE{actqt - Axt} t=O t=O t=O

for all {qt, xt} E Y, with equality if {qt, xt} = {q?, x4}. Then letting po = A 1at, we have shown that {q?, x? } maximizes

Z ftE{pqt - xt} t=O over Y, or that condition (8) of the definition of industry equilibrium is satisfied.


To show that (7) is satisfied (almost everywhere) by {q?, xo, po}, observe that if {qu, x0} maximizes S over all elements of X satisfying (15), it also maximizes S over the subset of X on which xt = 0 whenever xo = 0. Over this set, (16) (and the fact that xo is clearly 0 whenever yt = 0) implies that the constraint (15) is equivalent to

00 00

Z ftE{pqt - xt < E Z tE{poqo - X0}. t-O t=O Since s is strictly increasing in q, this constraint will be binding, so that {q?} maximizes

coo Z ftE{s(qt, ut) - p0qt]}. t=O

Then since qo > 0, we have

p?sl(q? ut) = D(q? ut),

which proves that (7) is satisfied (almost everywhere) and completes the proof of Lemma 1.

We now turn to the study of the problem of maximizing S subject to (1) and (2), the solution to which will be, by the Lemma just proved, the unique industry equilibrium. This will be done by studying the functional equation (9), which is related to the problem of maximizing S by the principle of optimality, a version of which is utilized in Lemma 4, below.

To show the existence and uniqueness of a solution to (9) and to obtain some of its properties, we utilize the method of successive approximation as applied in [2]. As a convenient device in this argument, we employ the operator T, taking bounded Baire functions on (0, oo) x E into the same set of functions, defined by

(17) Tf(k,u) = sup {s(k,u) - x + f,f( kh(4 ,z)P(dz,u)}.

Then, clearly, solutions to (9) are coincident with solutions to Tf = f. The relevant properties of T are given by the next lemma.

LEMMA 2: (i) Iff is non-decreasing in k, so is Tf; (ii) iff is concave in k, so is Tf; (iii) Tf = f has a unique solution, f*; and (iv) for any g, lim 0 T g = f*.

PROOF: To prove (i), observe that if f is non-decreasing, s(k, u) and f f [kh(x/k), z]p(dz, u) are non-decreasing functions of k for all fixed x.

To prove (ii), note first that the concavity off in k implies that f is continuous in k on (0, oo), so that the expression in brackets on the right of (7) is continuous in k. Further, this expression is finite at x = 0 and tends to - oo as x becomes large (since f is bounded). Hence, for any k > 0, Tf(k, u) is attained by some x >? 0.


Let k? and k' be on (0, oo), let k' be a convex combination, and let x? and x' attain Tf(kV, u) and Tf(k', u) respectively. Then:

Tf(k0, u) = s(k0, u) + sup {- x + 1B Jf [koh(x/k'), z]p(dz, u)} > Os(k0, u) - Oxo + 0/3 jf[k?h(x?/ko), z]p(dz, u)

+ (1 - O)s(k',u) -(1- O)x'

+ (1 - 0)1 f [k'h(x'l/k'), z]p(dz, u) = OTf(k0, u) + (1 - 0)Tf(k, u).

To prove (iii) and (iv), observe that T is monotone (f ) g for all (k, u) implies Tf > Tg) and if a is any constant, T(f + a) = Tf + af. By Theorem 5 of [3], these two facts imply (iii) and (iv).

Lemma 2 leads directly to a third lemma.

LEMMA 3: The functional equation (9) is satisfied by a unique bounded Baire function v(k, u) on (0, oo) x E. The function v(k, u) is continuous, non-decreasing and concave in kfor anyfixed u. For any (k, u), v(k, u) is attained by a unique investment rate x(k, u) and x(k, u) is a Baire function, continuous in k.

PROOF: Let g(k, u) = s(k, u). Then by Lemma 2, limn,,, Tg =f*, where f* is the unique bounded solution to (9). Let v(k, u) = f*. Since s(k, u) is increasing in both arguments and strictly concave in k, the limit function v is non-decreasing and concave in k. Since v is concave in k for fixed u, it is continuous in k on (0, co).

It follows from these facts that the expression in brackets on the right side of (9) is a continuous, strictly concave function of x, positive at x = 0 and negative for x sufficiently large, so that it attains its supremum on [0, oo) at a unique x, depending on k and u. This defines the policy function x(k, u).

To show that x(k, u) is a Baire function, we must show that for all xo, S(x0) = {(k, u) E (0, cc) x E: x(k, u) < xo}

is a Borel set. For x? < 0, this is trivial. Now denote the expression in brackets on the right of (9) by H(x, k, u), and for x? > 0 and e > 0, define

S(x?, s) = {(k, u) E (0, cc) x E: H(x?, k, u) > H(xo + s, k, u)}. Since H is a Baire function, - H(x?, k, u) + H(xo + s, k, u) is a Baire function on (0, cc) x E, for each e > 0. Hence S(x?, s) is a Borel set. Let {Enj be a decreasing sequence tending to 0 as n tends to infinity. Then

00 n S(xo? En)

is a Borel set, and since this set equals (using the concavity of H in x) the set S(x?),


x(k, u) is a Baire function. Since the expression in brackets varies continuously with k, x(k, u) is continuous in k.'0

The function x(k, u) defines, for given (ko, u0), a sequence of Baire functions tkt, x,} as given by (10) and (11). The next lemma shows that this sequence is the (essentially) unique optimal policy.

LEMMA 4: For anyfixed (ko, uo) E (0, oo) x E, the sequence {k?, xo} defined by (10) and (11) is an element of Y, and this policy maximizes S over Y. Further, it is the unique optimal policy in the sense that any other element of Y yielding the same value of S differsfrom {k?, x?} at most over sets of P, measure zero.

PROOF: From Lemma 3 and the continuity of h, the terms of the sequence {k?, x?} are Baire functions, and they clearly satisfy (1) and (2). To show that {k?, x??} E Y, then, we need only show that ko and xo are bounded, uniformly in (u1, . . . , ut, t). Since the policy xt = 0 for all t is always feasible, we have v(k, u) > 0 for all k > 0. Then from (9),

-x(k, u) + f{vLkh x(k u) z]p(dz, u) > 13{v[kh(O), z]p(dz, u) > 0 for all (k, u), k > 0. Let B be a bound for v so that the term on the left, above, is less than

-x(k, u) + /3B. Hence x(k, u) is bounded from above by /3B. Then if kt > (13B/3), kt + 1 < kt, so that max[(fiB/3), ko] is an upper bound for kt.

To show that {k?, x? } is optimal, it is sufficient to show that it yields the return v(ko, u0) [3, Theorem 6, part (f)]. Define the operator Fx, taking bounded Baire functions on (0, oo) x E into the same set of functions, by

Fxf(k, u) = s(k, u) - x(k, u) + /3 { f kh x(k u) , zlp(dz, u).

The functional equation Fxf = f has a unique solution, f* [3, Theorem 5], and this solution, evaluated at (ko, u0), gives the value of S under the policy {k?, x?}. Since Fxv = v, v = f*, which proves that {k?, x?} is optimal.

To prove uniqueness, let {k 1, xt } yield the same value to S as {k?, xo }, and let At be the Borel subset of Et on which x 1 and xo differ. Let t' be the first period in which the probability of At is non-zero. Then the returns from the policies differ by

ft'E{-xt, + v[ko4h( ),ut+i] + xl - v [ko4h{),ut+ 1 > 0,

10 A theorem in [4, p. 19], suitably specialized to this application, states that iff(x, y) is continuous, the function g(y) = maxXEs f(x, y), where S is compact, is continuous. Further, if g(y) is attained at a unique x(y) for each y, x(y) is also continuous. This fact will be used at various points below, without reference.


where the expectation is taken with respect to P, + 1 . This contradicts the assumed optimality of {kl, xl }, and completes the proof of the Lemma.

The proof of Theorem 1 now follows directly from Lemmas 1, 3, and 4. In the next two sections, we pursue the study of the unique industry equilibrium as given by (10H13).

6. LONG RUN EQUILIBRIUM WITH INDEPENDENT ERRORS Equations (10) and (11) and p( , * ) determine a Markov process {kt, ut} taking

values on (0, oo) x E which governs the development of kt, xt, and u, through time, starting from a given (ko, uo). To determine the long run characteristics of this process, it will be necessary to restrict the {ut} process further. In the present section, we treat the special but interesting case where ut and us are independent for s =A t, or where the transition function p(z, u) does not depend on u.

In this case, the functional equation (9) becomes

(18) v(k, u) = max {s(k, u) - x + 3 {vLkh( ),z1p(dz)}, from which it is clear that the optimal investment rate, x(k, u), will not depend on u. It follows from (10) and (11) that the time path of capital stock will be deterministic, following the difference equation

(19) kt+1 = kh( xkt))' where x(kt) is the unique investment rate attaining v(kt, u). Hence we turn our attention to the existence, uniqueness, and stability of stationary solutions to (19), with results summarized in Theorem 2, below.

A capital stock kc > 0 will be a stationary solution to (19) if and only if it is a solution to

(20) x(k) = bk, since h(b) = 1. The solutions to (20) are described in the next lemma.

LEMMA 5: Equation (20) has a solution kc > 0 if and only if (21) I D(O, u)p(du) > 3 +

r

A positive stationary solution, if it exists, must satisfy

(22) X D(k, u)p(du) = - + r

so there is at most one positive solution to (20). PROOF: We first show that any solution to (20) satisfies (22). From (18),

v(k, u) > s(k, u) - bk + ,B v(k, z)p(dz)


(since x = bk is always feasible) for all (k, u) with equality if and only if k satisfies (20). Then taking the mean of both sides with respect to p(u) and collecting terms: (23) E{v(k, u)} ) 1 1 / [E{s(k, u)} - bk] with equality if and only if k satisfies (20). If kc satisfies (20), (18) implies

- 3kC + fE{v(kc, u)} > - x + JJE{v(kch(x/kc))} for all x > 0. Now applying (23), which holds with equality at kc,

- bkC + 1 -

A [E{s(kc, u)} - bkc] >-x + [E{s(kch(x/kc)} - 3kch(x/kc)1,

for x > 0. At x = bkc, this inequality holds with equality. At any other value of x, the right side takes on a smaller value, which is to say that bkc maximizes the expression on the right. Then the first order condition

r E{sl(kc, u)} = E{D(kc, u)} = 3 + h(3)

is satisfied. If (21) does not hold, no kc > 0 satisfies this condition, so the necessity of (21) is proved. Further, we have proved that positive stationary solutions satisfy (22) and, since D is strictly decreasing in k, that there is at most one such solution.

To show that (21) is sufficient for the existence of a positive solution to (20), we must rule out the possibilities that x(k) > bk or x(k) < bk for all k > 0. To rule out the former possibility, recall from Lemma 4 that x(k) is bounded, so that for k sufficiently large, x(k) < bk. Suppose, contrary to the Lemma, that (22) has a positive solution kc and that x(k) < bk for all k > 0. Define the function z(k) by

z(k) = max E{s(k, u)} -bk so that from (23), z(k) < E{v(k, u)} for all k > 0. Now, define the operator Fx on bounded continuous functions on (0, oo) by

Fxy(k) = E{s(k, u)} - x(k) + flyLkh< k))1. It is readily verified that for any y, z in the domain of Fx, IFxy -FzII F ,BIy - zil. Also, FxE{v(k, u)} = E{v(k, u)} where v is the solution to (9), and for any y, (24) lim Fy =E{v(k, u)}.

n0o

We next show that for z(k) as defined above, Fxz(k) < z(k) for all k < kc. We have, directly from the definitions of z and Fx,

Fxz- 1 _ flE{s[kh ( ))u]}- E{s(k, u)}]

(25) _ 2 bkh x(k) - 1i + bk - x(k).


The strict concavity of s and h implies

(26) s[kh( k ),u - x(k,u) < D(k,u)kLh(') - Ii and

(27) h(' ( -1 < h'(Lx) - Combining (25), (26), (27), and (22) then gives, for k < kc, (28) Fxz < z. One may also verify that y(k) < z(k) over the interval (0, kcj implies FAy(k) < Fxz(k) on this interval. Thus (24) and (28) together imply

E{v(k, u)} = lim F'z(k) < z(k), k E (0, kc], nx-+Xo

which contradicts (23). This completes the proof of Lemma 5.

It can be shown that v(k) E{v(k, u)} is differentiable for k > 0.11 Then from (18), x(k) must satisfy

(29 -1 + flv Lkh(I x)k) hi [x(k)1 < ( ) : [ ( ~~k )][k] with equality if x(k) > 0. Inspection of (29) reveals that x(k)/k is a strictly decreasing function of k where x(k) > 0, and that kh(x(k)/k) is a strictly increasing function of k. Hence if (20) has a solution, kc > 0. Then the results of this section may be summarized in this theorem.

THEOREM 2: Under the hypothesis of independence of the {ut} process, there are two possibilitiesfor the behavior of the optimal capital stock, k,. If (21) holds, and if ko > 0, k, will converge monotonically to the stationary value kc, given implicitly by (22). If (21)fails to hold, or if ko = 0, capital will converge monotonically to zero.

The marginal condition (22), which is satisfied by the long-run capital stock, has a familiar interpretation. The left side of (22) is the expected marginal value product of capital. Since in our model the marginal physical product is unity, this becomes simply expected output price. The right side of (22) is a rental price or user cost of capital, equal to the sum of a depreciation cost term, 6, and a capital or interest cost term, (r/h'(6)).

While capital and investment are varying deterministically as described in Theorem 2, output each period is supplied to the market inelastically, in a quantity

l If we let gn(k) = E{T's(k, u)} and aj(k) = gn(k) - g,- 1(k), it can be shown that (i) an(k) < fnB1, (ii) aj(O) = 0, (iii) an(k) > 0, and (iv) la"(k)l < B2 provided h"(k) is bounded, for k > k* > 0. Given these conditions, it follows an(k) < (fnl'2/2)B1'2B2 proving g'(k) converges uniformly for k > k*. Since gn(k) converges to v(k) = Ev(k, u), v(k) is differentiable for all k > k*. Since this result holds for all k* > 0, v'(k) exists for all k > 0.


given by the historically determined capacity of the industry. Fluctuations in demand then affect price only, in a manner which can be computed from the demand function (7). (It may be noted that if variable inputs were introduced, the short-run supply would be upward sloping but not vertical, so that both price and output would vary with demand shifts. Nevertheless, in this case as well the capital- investment path will be deterministic.)

While the case of independent errors discussed in this section may appear to be an unlikely specialization of our general model, it is interesting to note that it corresponds perhaps more closely than any other analytical model of the industry to the familiar geometrical dichotomy between short-run and long-run supply. In the short-run case, capacity is fixed and price and output are determined entirely by the current demand function and the short-run supply function. In determining the long-run equilibrium, on the other hand, demand fluctuations play no role, with equilibrium capacity determined entirely by average, or normal, demand.

7. LONG RUN EQUILIBRIUM WITH SERIALLY DEPENDENT ERRORS

In the case of a serially independent {u,} process, as studied in the preceding section, a demand shift in period t results in a windfall gain (or loss) for firms in that period, but yields no information about what the state of demand in future periods is likely to be. As a result, the current state of demand has no effect on investment policy, and the capital stock of the industry follows a deterministic difference equation. In this section, we shall drop the assumption of independent demand shifts, replacing it with an assumption that these shifts are positively correlated in a particular way, so that, for example, an upward shift in demand not only increases price and profit this period but increases the probability that price will continue to be higher than average over the near future.

To study the case of dependent errors, it will be necessary to impose some additional restrictions on the {u,j process, with the general aim of assuring that the distribution of (ks, u,) will converge to a stationary distribution which is independent of the initial state of the industry, (ko, uo). First, we require, for all v E E and all non-degenerate intervals A c E,

(30) { p(du, v) > 0, so that for any u, and any non-degenerate subinterval of E, u,, 1 will fall in that subinterval with positive probability. Similarly, it assures that ut,, will fall in that subinterval with positive probability, or that

(31) pr [ut+s E Alut = uo] > 0 for all uo E E and all non-degenerate subintervals A of E. Assumption (30) does not preclude the possibility that the left side of (31) may approach 0 as s approaches infinity. We shall rule out this possibility explicitly by adding

(32) lim pr[ut+ sE Alut = uo] > 0 s-4 00


for all u0, A as above, where the limiting value does not depend on the value of u0. That is, ut has a limiting distribution which does not depend on the initial value u0 and which assigns positive probability to all non-degenerate intervals of E. It follows from (31) and (32) that for any fixed A the terms of the sequence (31), s = 1, 2,... ., are uniformly bounded away from zero.

We also wish to insure that the relation between current and future demands is always positive (so that a high demand this period always signals high demand for the future) by requiring that (33) pr {ut+ 1 ) xlutj be a strictly increasing function of ut for all x E E and that

[lim pr{ut+, A xlut = 1, (34) JUt --o0 ( lim pr {ut+1 > xlutj = 0,

tUt --*-0

for all x E E. An example of a process satisfying the restrictions (30), (32), (33), and (34) is:

ut+1 = Put + St

where 0 < p < 1, and where {Is} is a sequence of independent, identically distri- buted normal random variables. Finally, in addition to the assumption (Section 2) that D(k, u) (and hence s(k, u) also) is an increasing function of u, we add the restrictions that the limits

(35) s(k) = lim s(k, u), U+ O0

(36) s(k) = lim s(k, u), U+ 00

exist, and convergence is uniform in k. Our first task, given these restrictions , is to decompose the set of possible states

of the (kt, ut) process, the set (0, oo) x E, into transient sets (sets which cannot be entered, and which will be departed with probability ultimately approaching 1) and ergodic sets (sets which once entered cannot be departed, and which contain no transient subsets). To do this, we develop some additional notation and prove some preliminary lemmas.

Since fv(k, z)p(dz, u) is a bounded, increasing (by (33)) function of u, the limits

(37) v-(k) = lim J'v(k, z)p(dx, u),

(38) v(k) = lim f v(k, z)p(dz, u), U-oo 0

exist for all fixed k. We wish to show that convergence is uniform in k and to exhibit functional equations solved by v(k) and v(k). To this end, we first introduce another lemma.


LEMMA 6: Let f(k, u) be a bounded Baire function on (0, oo) x E, such that lim f(k, u) = f(k) and lim f(k, u) = f(k)

U-f 00 U-- c

exist, and convergence is uniform. Then

lim j f(k, z)p(dx, u) = f(k) and lim J f(k, z)p(dx, u) = (k), U-c Jo U-4 - oo

and convergence is uniform.

PROOF: Let B be a bound for f. Then for x E E,

f(k) - f(k, z)p(dz, u)I

- | g [f(k) - f(k, z)]p(dz, u) + [f(k) - f(k, z)]p(dz, u) rx 00

< 2BJ' p(dz,u) + sup If(k) - f(k, z)I p(dz, u). ze(x, o) x

For x sufficiently large, If(k) - f(k, z)I is arbitrarily small, uniformly in k, while fJ?p(dz, u) is bounded by 1. For any fixed x, however large, the first term on the right is made arbitrarily small by choosing u sufficiently large, applying (34). The proof for f(k) is similar.

As an application of Lemma 6, we have the next lemma.

LEMMA 7: The function v(k), defined in (37), satisfies the functional equation

(39) v(k) = sup {(k) - x + vT [kh )1} and an analogous functional equation is satisfied by v(k).

PROOF: We first show that

(40) lim v(k, u) = v(k) u- 00

uniformly in k. Define the operator T as in Section 3, and let O(k, u) = 0 on (0, oo) x E. Let On(k, u) = TnO(k, u). Then, by Lemma 3, lMn, TnO = v(k, u), uniformly in k and u. It will then be sufficient to show that On(k, u) converges uniformly to the limit On(k) as u -+ oo, for all n. For n = 1, the proposition is true, since 01(k, u) = s(k, u), and s(k, u) converges uniformly to s(k) by (35). Suppose O(k, u) converges uniformly to O(k). By the definition of the operator T,

On+ 1 (k, u) = supO { s(k, u) -x + ?n kh X , z p(dz, u)


Let

On+1(k) = s(k) + sup {-x + flnLkh(k)]}. Since O,(k) > On(k, u),

0 K On + l(k) - On +(k, u)

=U su {X + fl~khF~ x

x {> 00 [kh(

sup {-x + n J 0nLkh () z p(dz, u)} + s(k) -s(k, u) 3SUP OnLkh( x I kh()z1p(dz, u) + 0(k) - s(k, u).

By Lemma 6, the induction hypothesis, and (35), the term on the right is arbitrarily small, uniformly in kh(x/k), for u sufficiently large. Thus 0.+ 1(k, u) converges uniformly to On+ l(k) and limn . O(k). This completes the proof that the convergence in (40) is uniform.

Now v-() = lim sup s(k, u) - x + / fv[kh(x/k)]p(dz, u)} u-oo x3>0

By (35), (40), and Lemma 6, the expression in braces converges uniformly in k; thus, the supremum and limit operation can be interchanged to obtain (39). The proof for v(k) is similar

The limit functions v-(k) and v(k) share the properties of continuity, monotonicity, and concavity with f v(k, z)p(dz, u) regarded as a function of k for fixed u. Then, as in Section 4, v(k) and v(k) are attained at unique values x(k) and x(k) respectively, and these functions are continuous. Also as in Section 4, the functions (1/k)5-(k) and (1/k)x(k) are decreasing and the functions kh[(1/k)5-(k)] and kh[(1/k)x(k)] are increasing. Finally,

5-(k) = lim x(k, u) u-+ 00

and x(k) = lim x(k, u)

u-00

for all k. From (33), x(k, u) is an increasing function of u, so that x(k) > x(k) for all k such

that x(k) > 0. By an argument used to prove Lemma 4, 5-(k) < bk for k sufficiently large. Let k be the least positive solution to 5-(k) = bk (or 0 if there is no positive solution) and let k be the greatest positive solution to x(k) = k (or 0 if there is no solution). Then since x(k) < 5(k) if 5-(k) > 0, either k =k = 0, or 0 < k < k < oo.

We now prove another lemma.

LEMMA 8: (i) For all k > k, there exists ho < 1 (not dependent on k) such that the event h[(x(k, u)/k)] < ho has positive probability, for all t and all uo; (ii) for all


0 < k < k there exists h1 > I (not dependent on k) such that the event h[(x(k, u/k)] ) h1 has positive probability,for all t and all uo.

PROOF: We prove (i) only, since the proofs of the two parts are essentially the same. For any k, the event (1/k)x(k) < (1/k)x(k, u) < (1/2k)[5x(k) + x(k)] has positive probability, by (31) and the definitions of 5x(k) and x(k). For k > k, (1/k)x(k) < (1/k)x(k) = ,so that the occurrence of this event implies

-x(k, u) < - + - x(k) < . k 2 2k Letting ho = h[(6/2) + (1/2k)x(k)] < 1, (i) is proved.

Using Lemma 8, we can prove the next lemma.

LEMMA 9: The sets T1 = (0, k) x E and T2 = [k, so) x E are transient, and the set B = (k, k) x E contains all ergodic sets.

PROOF: Since {kt, ut} takes values in T1 u B u T2 (for ko > 0), it is sufficient to show that (i) if (k0, uO) 0 7T (i = 1, 2), (kt, u,) 0 Ti with probability 1 for all t = 1,2.. 2. and (ii) for all (ko, uo) e T1 u T2,

lim pr [(kt, ut) E B] = 1.

To prove (i), we show that pr [kt+ 1 > kl(kt, ut)] = 0 for (kt, ut) E T1 u B. For kt < k, we have

kt+ < kth(k x(kt)) < kh(15(k)) = k

since kh((1/k)x(k)) is an increasing function of k. A similar argument rules out passage into T1 from B u T2.

To prove (ii), let (ko, u1) E T2. Since x(ko, u) < bko for all u, we have.kt < ko for all t. Let ho < 1 be the number whose existence was established in Lemma 8. For some t*, kohl* < k. Hence if the event h[(x(k, u)/k)] < ho occurs t* times in t periods, kt < k. Since the probability that this event occurs is bounded away from zero, the probability that it will not occur t* times in t periods goes to zero as t becomes large. A similar argument applies if (ko, uo) E T1.

Next, we have the final lemma.

LEMMA 10: The set B = (k, k) x E constitutes the single ergodic set.

PROOF: It has been shown (Lemma 9) that all ergodic sets are contained in B. To show that B is a single ergodic set, it is sufficient to show that for some t ) 1, all (ko, uo) E B, and all non-degenerate rectangular subsets A of B, that

pr [(kt, u) E AI(ko, u0)] > 0.


Let A = {(k,u)eB: ka < k < kb, Ua < U < Ub}, where k < ka < kb < k and Ua < Ub, Ua, Ub E E. Suppose, to be specific, that ko < ka. Then for any finite t one can choose 02 > 01 > 1 such that k00' = ka and k00' = kb, and limtO 01 = 1. By choosing t large enough, the event h[x(k, u)/k] > 01 has positive probability for all k < ka, so that the event

01 < h[x(k, u)/k] 1< 0 2 has positive probability. Hence the probability that this event will occur t consecu- tive times is positive, as is the probability that Ua < Ut < Ub for any u-1 Hence the lemma is proved.

Lemmas 9 and 10 accomplish the division of the set (0, oo) x E of possible states of the {k,, u,} process into two transient sets and a single ergodic set. In doing this, nothing has been assumed to assure that the ergodic set B is non-empty, or equivalently, that x(k) = bk has a positive solution k. The following theorems characterize the long-run behavior of the system under the hypothesis that B is non-empty. We return to the questions of verifying this hypothesis, and of charac- terizing the system when it is false, below.

THEOREM 3: If B is not empty, for all (k, u) and all (ko, uo), the limit lim pr {kt < k, ut < ul(ko, uo)} = P(k, u) t+ oo

exists, and does not depend on (ko uo). The function P(k, u) is a probability distribution on (0, oo) x E, assigning probability zero to all subsets of T1 u T2 and positive probability to all Borel subsets of B with positive area.

PROOF: See [6, Theorem 5.7, p. 214, and surrounding text].

Since under the limiting distribution, kt lies on a bounded interval with probability 1, the mean of k, k* say, exists and is positive. Then we have the following "stability" theorem.

THEOREM 4: If B is not empty, then for any initial state (ko0 uo), l T

lim T E kt = k*

with probability 1.

PROOF: The result follows from [6, Theorem 6.2, p. 220].

To verify the hypotheses of Theorems 3 and 4, we wish to determine when the equation x(k) = bk has a positive solution. This question is analogous to the question of the existence of a positive solution to x(k) = 3(k) studied (and solved) in the preceding section. We have the following theorem.


THEOREM 5: The ergodic set E is non-empty (x(k) = bk has a positive solution) if and only if

r (41) lim D(O, u) > 3 +

r

PROOF: The proof follows that of Lemma 5, with the operator J f(u)p(du) on functions f replaced by limu,b f(u). This replacement is justified by Lemma 7.

If (41) fails to hold, capital stock in the industry will go to zero with probability one, since x(k,, u,) < bk, for all (ks, u,). Thus Theorems 3, 4, and 5 provide a complete description of the long-run behavior of the industry under the assumptions of this section.

8. CONCLUSIONS

The object of this paper has been the extension of "cost-of-adjustment" type investment theory to situations involving demand uncertainty. In doing so, we have tried to go beyond formulations of the "price expectations affect supply, which in turn affects actual price" variety, to consider the simultaneous determination of anticipated and actual prices. This involves studying the determination of industry equilibrium, in addition to the individual firm's optimizing behavior, a step which radically alters the nature of the problem.

Generally, the equilibrium behavior of capital stock, output, and price through time is similar to the certainty case studied in [14]: the interplay of shifting demand and the costs of varying capacity leads to a difference equation in capital stock. The stationary character of the demand shifts leads capital stock to settle down, either with certainty or "on average," to a long-run equilibrium level, determined by interest rates, adjustment costs, and average demand levels.

An interesting feature of our theory is the role played by securities prices in informing firms of the market (not "shadow") price placed on additions to capital stock. We have found (Section 3) that securities prices as a variable in a firm level investment function have a much stronger justification than simply as a "proxy" variable for future demand.

Carnegie-Mellon University

Manuscript received June, 1969.

REFERENCES

[1] ARROW, K. J.: "The Role of Securities in the Optimal Allocation of Risk Bearing," Review of Economic Studies, 31 (1964), 91-96.

[2] BELLMAN, RICHARD: Dynamic Programming. Princeton: Princeton University Press, 1957. [3] BLACKWELL, DAVID: "Discounted Dynamic Programming," Annals of Mathematical Statistics, 36

(1965), 226-235. [41 DEBREU, GERARD: Theory of Value. New York: Wiley, 1959. [5] "Valuation Equilibrium and Pareto Optimum," Proceedings of the National Academy of

Science, 40 (1954), 588-592.


[6] DOOB, J. L.: Stochastic Processes. New York: Wiley, 1953. [71 EISNER, ROBERT, AND ROBERT STROTZ: "Determinants of Business Investment," (with biblio-

graphy by G. R. Post) in Commission on Money and Credit, Impacts of Monetary Policy. Englewood Cliffs: Prentice-Hall, 1963, pp. 59-337.

[8] GOULD, J. P.: "Adjustment Costs in the Theory of Investment of the Firm," Review of Economic Studies, 35 (1968), 47-55.

[9] GRUNFELD, YEHUDA: "The Determinants of Corporate Investment," in A. C. Harberger (Editor), The Demandfor Durable Goods. Chicago: University of Chicago Press, 1960.

[10] HAAVELMO, TRYGVE: A Study of the Theory of Investment. Chicago: University of Chicago Press, 1960.

[11] HIRSHLEIFER, J.: "Investment Decision under Uncertainty: Choice Theoretic Approaches," Quarterly Journal of Economics, 79 (1965).

[12] JORGENSON, DALE W.: "Capital Theory and Investment Behavior," American Economic Review, 53 (1963), 247-259.

[13] LUCAS, ROBERT E., JR.: "Optimal Investment Policy and the Flexible Accelerator," International Economic Review, 8 (1967), 78-85.

[141 "Adjustment Costs and the Theory of Supply," Journal of Political Economy, 75 (1967), 321-334.

[151 MUTH, JoHN F.: "Rational Expectations and the Theory of Price Movements," Econometrica, 29 (1961), 315-335.

[161 PASHIGIAN, B. PETER: "Supply Uncertainty and the Theory of a Competitive Industry," University of Chicago working paper, 1968.

[17] PRESCOTT, EDWARD C., AND ROBERT E. LUCAS, JR.: "A Note on Price Systems in Infinite Dimen- sional Spaces," International Economic Review, forthcoming.

[18] SCHRAMM, RICHARD: "Optimal Adjustment of Factors of Production and the Study of Investment Behavior," Carnegie Institute of Technology, doctoral dissertation, 1966.

[19] TREADWAY, ARTHUR: "Rational Entrepreneurial Behavior and the Dynamics of Investment," University of Chicago, doctoral dissertation, 1967.

Article Contentsp. 659p. 660p. 661p. 662p. 663p. 664p. 665p. 666p. 667p. 668p. 669p. 670p. 671p. 672p. 673p. 674p. 675p. 676p. 677p. 678p. 679p. 680p. 681

Issue Table of ContentsEconometrica: Journal of the Econometric Society, Vol. 39, No. 5, Sep., 1971Front Matter [pp. 865 - 865]Investment Under Uncertainty [pp. 659 - 681]Congestion Tolls for Commercial Airports [pp. 683 - 694]CRESH Production Functions [pp. 695 - 712]An Equivalence Theorem for the Core of an Economy Whose Atoms Are Not "Too" Big [pp. 713 - 721]Comparison of k-Class Estimators When the Disturbances Are Small [pp. 723 - 737]Determinants of Negotiated Wage Increases: An Empirical Analysis [pp. 739 - 750]The Identification Problem for Multiple Equation Systems with Moving Average Errors [pp. 751 - 765]A Nonconvex Control Problem for the Competitive Firm [pp. 767 - 772]Development Strategy for a Medium-Sized Economy [pp. 773 - 795]Random Walk of Stock Prices: A Test of the Variance-Time Function [pp. 797 - 812]Estimation of Net Social Benefits from Outdoor Recreation [pp. 813 - 827]Some Statistical Models for Limited Dependent Variables with Application to the Demand for Durable Goods [pp. 829 - 844]The Rational Multivariate Flexible Accelerator [pp. 845 - 855]Notes and CommentsThe Strict Convexity of the Transformation Surface in Case of Linear Homogeneous Production Functions: A General Case [pp. 857 - 859]Non-Linear Estimates of the Liquidity Trap [pp. 861 - 864]

Book Reviewsuntitled [p. 867]untitled [p. 868]untitled [pp. 868 - 869]untitled [pp. 869 - 871]untitled [p. 871]untitled [p. 872]

Accepted Manuscripts [pp. 873 - 874]Membership List [p. 874]European Meeting of the Econometric Society [pp. 875 - 876]La Confrence Europenne de la Socit d'conometrie [pp. 877 - 878]News Notes [p. 879]Back Matter

Documents

Investment under uncertainty - Robert Emerson Lucas, Jr. y Edward C. Prescott